What is the Turing test?

Source: Internet
Author: User
Keywords algorithm artificial intelligence testing natural language processing turing test

The two questions in the previous section, "How do you determine intelligence?" and "Is the animal smart?" have been resolved. The answer to the second question is not necessarily simple “yes” or “no”—some people are smarter than others, and some animals are smarter than others. Machine intelligence has also encountered the same problem.

Alan Turing seeks an actionable way to answer intelligent questions, separating the functions (things that intelligence can do) from the implementation (how to implement intelligence).

Abstraction is a strategy that ignores the implementation of objects or concepts (such as internal work), so that you can get a clearer picture of artifacts and their relationship to the outside world. In other words, you can think of this object as a black box, focusing only on the input and output of the object.

Often, abstraction is a useful and necessary tool. For example, if you want to learn how to drive, it might be a good idea to think of the car as a black box. Instead of trying to learn about automatic transmissions and powertrains from the start, you can focus on system inputs such as accelerator pedals, brakes, turn signals, and outputs such as forward, stop, left and right. The course of the data structure also uses abstraction, so if you want to understand the behavior of the stack, you can focus on the basic stack operations, such as pop (pop-up one) and push (insert one), without having to fall into the details of how to construct a list ( For example, whether to use a linear linked list or a circular linked list, or to use a linked linked list or to allocate space continuously).

Turing test definition

Alan Turing presented two simulation games. In a simulation game, a person or entity behaves as if it were another person. In the first simulation game, a person in a room with a curtain in the center, one on each side of the curtain, one side of the person (called the inquirer), must determine whether the other side is a man or woman. The inquirer (its gender does not matter) does this by asking a series of questions. The game assumes that men may lie in his answers, while women are always honest. In order to make it impossible for the inquirer to determine the gender from the voice, the communication is carried out by means of a computer instead of a speech, as shown in Fig. 1.3. If there is a man on the other side of the curtain and he successfully deceives the inquirer, then he wins.

The original form of the Turing test was that a man and a woman were sitting behind the curtains and the inquirer had to correctly identify the gender (Turing could be inspired by the popular games of that era and invented the test. This game also motivated him. Perform a machine intelligence test). As Erich Fromm wrote [8]: equality between men and women, but not necessarily the same. For example, people of different genders have different knowledge about colors and flowers, and spend less time shopping. What is the relationship between men and women and intelligence? Turing believes that there may be different types of thinking, and it is important to understand and tolerate these differences. Figure 1.4 shows the second version of the Turing test.

The second game is more suitable for artificial intelligence research. The inquirer is still in the room with the curtain. This time, the back of the curtain may be a computer or a person. The machine here plays the role of a male, occasionally lying, but the person is always honest. The inquirer asks questions and then evaluates the answers to determine whether he is communicating with people or communicating with the machine. If the computer successfully deceives the interrogator, then it passes the Turing test and is therefore considered to be intelligent.

It is well known that machines perform many times faster than humans when performing arithmetic calculations. If the "person" behind the curtain can get the result of the Taylor series approximation of the trigonometric function within a few microseconds, then it is easy to discern that the computer behind the curtain is not a person.

Naturally, the chances that a computer can successfully deceive an interrogator in any Turing test is very small. In order to get an effective intelligent "barometer", this test has to be performed many times. Similarly, in the original version of this Turing test, both the person and the computer were behind the curtain, and the inquirer had to correctly identify them.

Turing test

No computer system passed the Turing test. However, in 1990, philanthropist Hugh Gene Loebner hosted a competition designed to implement the Turing test. The first computer to pass the Turing test will be awarded a gold medal and a $100,000 Robner bonus. At the same time, the best performing computer in the competition each year will be awarded a bronze medal and a prize of approximately $2000.

What questions would you ask in the Turing test? Consider the following example:

· (1 000 017)? What is it? Calculations like this may not be a good idea. Remember, the computer is trying to trick the inquirer. The computer may not respond within a fraction of a second, giving the correct answer, it may intentionally take longer, and may make mistakes because it "knows" that humans are unfamiliar with these calculations.

· What is the current weather situation? Suppose the computer may not look out the window, so you might try to ask the weather. However, the computer is usually connected to the World Wide Web, so it is also connected to the weather website before answering.

· Are you afraid of death? Because computers are hard to disguise people's emotions, you may ask this question or other similar questions: "How does black feel for you?" or "How does it feel to fall in love?" But remember, you are trying now. Judging intelligence, human emotions may not be effective intelligent "barometers."

Turing expected that there would be many people who opposed the idea of ??"machine intelligence" that he proposed in his original paper, one of which was the so-called "ostrich policy opposition." People believe that the ability to think makes people become the spirit of all things. Admit that computers can think, which may challenge this lofty habitat that is only enjoyed by humans.

Many people believe that it is the human soul that allows people to think. If we create machines with this ability, we will usurp the authority of God. Turing refuted this view, suggesting that people are merely waiting to wait for containers with soul endowments to carry out the "God" will. Finally, we refer to the opposition of Lady Lovelace (she is often referred to as the first computer programmer in the literature).

When commenting on the analytical engine, she said with ease: "This machine alone can't surprise us." She reiterated the belief of many people that a computer cannot perform any unpreprogrammed activities. Turing opposed this opinion, saying that the machine always surprised him. He insists that proponents of such objections agree that human intelligence can immediately infer all the consequences of a given fact or action. Turing's original paper referred to these readers when collecting the above objections and other objections.

Turing test controversy and criticism

Ned Block believes that English text is encoded in ASCII, in other words, a series of 0s and 1s in the computer. Therefore, a specific Turing test, which is a series of questions and answers, can be stored as a very large number. For example, suppose the length of the Turing test has an upper limit. In the test, the first three characters starting with "Are you afraid of dying?" are stored as binary numbers.

Assuming that the typical Turing test lasts for an hour, during which the tester raises about 50 questions and gets 50 answers, the binary number corresponding to the test should be very long. Now, let's assume that there is a large database that stores all of the Turing tests. These Turing tests contain 50 or fewer questions that have reasonable answers.

Then, the computer can pass the test by looking up the table. Of course, a computer system capable of handling such a large amount of data does not yet exist. However, if the computer passes the Turing test, Block asks: "Do you think such a machine is smart? Do you feel comfortable?" In other words, Block’s criticism is that the Turing test can be done mechanically. Not smart to pass the Turing test.

John Searle's criticism of the Turing test is more fundamental. Imagine that the inquirer asked the question as expected - but this time it was in Chinese. The other person in the room didn't know Chinese, but had a detailed rule book. Although the Chinese question is presented in scribbled handwriting, people in the room will refer to the rules manual, process Chinese characters according to the rules, and write the answers in Chinese.

The interrogator received an answer to a grammatically correct and semantically plausible question. Does this mean that people in the room know Chinese? If your answer is "no", then is the combination of human and Chinese rulebooks fluent in Chinese? The answer is still “no”—the people in the room are not learning or understanding Chinese, but just dealing with symbols. Similarly, computers run programs that receive, process, and use symbolic responses without having to learn or understand what the symbol itself means.

Searle also asked us to imagine a scenario where a single person holds a rule book: in a gym, people pass notes to each other. When a person receives such a note, the rule book will determine whether the person should generate an output or simply pass information to another person in the stadium.

Now, where is the knowledge of Chinese? Belong to all, or belong to the gym?

Think about the last example. Depicting the brains of a person who knows Chinese well, as shown in Figure 1.8. This person can receive questions in Chinese and accurately explain and answer in Chinese.

Similarly, where does Chinese knowledge exist? Is it present in a single neuron or is it in a collection of these neurons? (It must exist somewhere!) The key to Block and Searle's criticism of the Turing test is that the Turing test is only viewed from the outside and does not provide insight into the internal state of an entity. In other words, we should not expect to learn something new about intelligence by treating smart agents (people or machines) as black boxes. However, this is not always true. In the 19th century, physicist Ernest Rutherford correctly inferred the internal state of matter by bombarding gold foil with alpha particles—it consisted mainly of white space.

He predicted that these high-energy particles would either pass through the gold foil or be slightly deflected. The result is consistent with his theory of atomic orbital: the atom consists of a dense core surrounded by orbital electrons. This is our current atomic model, and many people who have studied high school chemistry are very familiar with it. Rutherford successfully understood the internal state of the atom through external observation.

In short, it is difficult to define intelligence. It is because of the difficulty of defining intelligence and determining whether the "agent" has this property, so Turing developed the Turing test. In the paper, he implicitly pointed out that any agent that can pass the Turing test must have "brain ability" to deal with any reasonable intelligent challenge equivalent to the human level accepted in the general sense.

Related Article

Contact Us

The content source of this page is from Internet, which doesn't represent Alibaba Cloud's opinion; products and services mentioned on that page don't have any relationship with Alibaba Cloud. If the content of the page makes you feel confusing, please write us an email, we will handle the problem within 5 days after receiving your email.

If you find any instances of plagiarism from the community, please send an email to: info-contact@alibabacloud.com and provide relevant evidence. A staff member will contact you within 5 working days.

A Free Trial That Lets You Build Big!

Start building with 50+ products and up to 12 months usage for Elastic Compute Service

  • Sales Support

    1 on 1 presale consultation

  • After-Sales Support

    24/7 Technical Support 6 Free Tickets per Quarter Faster Response

  • Alibaba Cloud offers highly flexible support services tailored to meet your exact needs.